Exploring the meaning of gender:
Evaluating and revising the Bem Sex-Role
Inventory (BSRI) for a Swedish research
context (BSRI-SE)
Roland S. Persson
INSIKT 1999:1
Vetenskapliga rapporter från HLKAbstract
This research evaluates the Bem Sex-‐Role Inventory (BSRI) for use in a Swedish setting, and in the process of so doing also compares the distribution of American and Swedish gender roles as elicited by the inventory. A sample of 118 individuals (48 males and 70 females) was used in order to arrive at norms better suited to a Swedish context than those provided by Bem (1974). Reliability and factor analyses were performed and a revised version of the inventory, with acceptable psychometric properties for the Swedish sample, was arrived at and termed BSRI-‐SE. Comparisons between the original American and the Swedish sample show that differences in the way that femininity, masculinity, androgyny and undifferentiated gender are distributed are not statistically significant. Raw score data for the BSRI-‐SE, as well as the scoring sheet, are provided for continued standardisation of the inventory. Methods for classification and further research are discussed.
____________________________________________________________ Field of study: Social psychology
Keywords: Gender, Androgyny, Masculinity, Femininity,
Psychological Measurement, Cross-‐cultural
Index
Introduction, 4
The Bem Sex-Role Inventory (BSRI), 8 The BSRI in a Swedish context, 17 The nature of the Swedish sample, 17 Reliability of the inventory scales, 20
Procedure for classification and standardisation, 27
The median split method, 28 The M-‐F Difference score, 29
Are Swedes masculine, feminine or androgynous? 32
Sex-‐typing in the sample subgroups, 35
Concluding remarks, 36
References, 40
Appendices
I Test sheet for the BSRI-‐SE
II Raw scores of the Swedish sample
Introduction
Few issues in recent years have been subject to so much debate, research, exploitation and politics as have gender roles, their meaning and social impact. While researchers agree that sex is biologically determined they do not agree on when, why and how psychological differentiation between gender roles occur. To argue that something is typically male or female is by no means theoretically straight-‐forward (Sternberg, 1993), and there is a great deal of variety across cultures of what is regarded as typically male or female behaviour. Although there are some universals with regard to gender, societies differ in the value attached to maintenance of differences between the sexes. Generally, males are more likely to be physically aggressive and express dominance over women rather than vice versa. Women are more likely in a global perspective to conform, defer, comply and submit to authority, especially if the authority is male (Segall, Dasen, Berry & Poortinga, 1990). Men and women’s self-‐perceptions are less stereotypical in more economically and socially developed countries, and it appears that as sexual-‐equality ideals spread, for example through feminist ideology, behavioural differences between the sexes diminish and prevailing stereotypes are increasingly being blurred (Williams & Best, 1989).
Gender is a multi-‐facetted phenomenon. Most would think of gender in terms of feminine and masculine, thus expressing that gender is confined to a dichotomy. But there is in fact no agreement amongst behavioural scientists on how many genders there are! Is it
feasible to have as many genders as there are sexes, namely two? Or are there also “nuances” of both, which should perhaps be regarded as separate genders? Note that by gender is meant the psychological array of behaviours, that to a large extent are socially determined, and that we have a tendency to identify as typical of either sex. In some cases there are biological causes that could possibly qualify the conceptualisation of more genders than the traditional two. Sex chromosome disorders like Turner’s Syndrome (individuals born with only one X chromosome), Klinefelter’s Syndrome (individuals born with three X chromosomes or two X chromosomes and one Y chromosome) or the XYY Syndrome are all genetic disorders, which have distinctive influences on an individual’s secondary sexual characteristics like body hair, breasts, body build and so on. With these disorders follow some differentiation in cognitive functioning (cf. Willerman & Cohen, 1990 for an overview). In addition, there are a number of different sexual behaviours that challenge the traditional division of gender restricted to the dichotomy of male and female. Transsexualism, for example, describes incongruity between gender identity and anatomical sex, the aetiology of which could possibly be due to endocrine factors rather than in various ways social learning (Hoenig & Kenna, 1974). Similarly, male transvestism could be understood as the co-‐existence of two selves: male and female (Larsson, 1997). Homosexuality is also in a sense a challenge to traditional gender roles (Bailey, 1996; Ross, 1985; Williams, 1996), particularly since recent research is increasingly showing that homosexuality may be genetically and/or hormonally determined (Ellis, 1996; LeVay, 1993). A majority of researchers tend to stress
social aspects instrumental to developing sexual behaviour (Howells, 1987). Money and Erhardt (1972), for example, reviewed a large number of studies on sex hormone anomalies and concluded that the most important factor in developing sex role and gender identity is the assigned sex of rearing. However, such research may well be in error if understanding homosexuality as merely a social construction of no or little evolutionary significance, as elegantly argued by McKnight (1997). Genetics and human physiology seem to play important, but hitherto often wilfully ignored, roles in the development of a homosexual identity. Murphy (1990) provocatively points out that “the incentive to discover the origins of homosexuality seems to belong to those who find homosexuality a pathological, sinful, immoral or criminal condition. At least on the basis of these views there is reason to try and understand the origins of homosexual behaviour if only to prevent and eliminate it. It is ordinarily some deficit which prompts medicine and the rest to reach for a causal explanation of behaviour” (p. 134).
Thus, on the basis of culture, and invariably at some level, also biological sex—which at times by no means is always straight forward—it may well be possible to construe gender roles in terms of a masculine male, a feminine male, a masculine woman, a feminine woman or any nuance of these.
Rudberg and Bjerrum-‐Nielsen (1994) point out, that “gender is still relevant [in modern society]. There is nothing in either the family’s gender socialisation or the history of modernity that suggests that gender should be ‘suspended’ as a psychological or social category; rather it seems that gender will survive in a new form” (p. 49-‐50). The
understanding of gender is a field of knowledge in flux not only because it intrigues social scientists due to its complexity, but also because the results of research on possible causes and effects in this context have strong political repercussions. It is historically true in the Western world that women, often at great risk, have struggled in order to obtain the same rights that by unquestioned tradition have been the privilege of men only. However, during the process by which such rights and privileges are increasingly recognised for both sexes, the biased function of rationalisation should also be recognised.
While it is politically essential to argue that men and women have equal rights in a democracy it is simultaneously paramount for science to investigate the nature of differentiation. But where the political agenda is set differences between gender roles would appear not to be a favoured object of discussion or debate—at least in a context of equal opportunities on the job market. Rather issues are focussed which may justify a certain political development. In such a situation it is far more opportune to focus how men and women are similar rather than different, since anything different from what is considered typically masculine not infrequently is interpreted as inferior. Although many stereotypes about gender differentiation are false—like for example the notion prevalent amongst men and school teachers, that women generally are believed to be less intelligent than men (Broverman, Vogel et al., 1972)—it is my impression that real differences often are underplayed in favour of making women more like men (cf. Lipman-‐ Blumen, 1989). Baumeister (1988), for example, argues that differences between sexes should not be studied at all, whereas McHugh, Koeske and Frieze (1986) suggest that gender differences
should only be reported under limited circumstances
This research targets the social differentiation of gender believing it to be essential. First, it sets out to evaluate the Bem Sex-‐Role Inventory (Bem, 1974) in a Swedish context, revising it to make it more akin to Swedish culture rather than American culture, and to provide norms by which the revised inventory may be compared. Similar efforts have been done in several Hispanic settings (e.g. Kaschak & Sharrat, 1983; Kranau, Green & Valencia-‐Weber, 1982). Some studies have used the BSRI for cross-‐cultural comparisons in Israel (Maloney, Wilkof & Dambrot, 1981) and Australia (Rowland, 1977), but both of these used the BSRI original norms on the slightly dubious assumption that the understanding of gender identity and behaviour is roughly the same as in the United States.
An instrument such as the BSRI, however, has value for research purposes, since gender consideration often is an important variable in Social Science research. Few instruments are developed or translated and standardised for a Swedish population. For example, the publishing branch of the Swedish Psychological Society in their 1997 catalogue lists 111 different tests and test batteries, which should be compared to The American Psychological Association’s estimate that some 20.000 tests are developed annually (APA, 1993).
Second, and a result of the process of evaluating the BSRI in a Swedish context, is an exploration into the meaning of Swedish gender roles. How does the Swedish sample compare to the American sample, which served as the basis for developing the American norms by which BSRI scores are compared? Are Swedes more or less masculine,
feminine, androgynous than Americans?
The Bem Sex-Role Inventory (BSRI)
The BSRI was developed by Stanford-‐psychologist Sandra Bem (1974; 1975; 1981) in an effort to accommodate the fact that there are more nuances to being masculine or feminine than the theoretical constructs at the time would allow to be expressed (e.g. The California Psychological Inventory by Gough, 1957). This sex-‐role dichotomy, Bem (1974) argues, “has served to obscure two very plausible hypotheses: first, that many individuals might be ‘androgynous’; that is, they may be both masculine and feminine ... depending on the situational appropriateness of these various behaviours; and conversely, that strongly sex-‐typed individuals might be seriously limited in the range of behaviours available to them as they move from situation to situation” (p. 155).
In Bem’s research effort masculinity and femininity are regarded as two independent dimensions rather than two ends of a single dimension. This conceptualisation of gender allows any individual to indicate whether he or she is high on both dimensions (and thus
androgynous), low on both dimensions (and therefore
undifferentiated) or high on one and low on the other (either feminine or masculine). Three scales were originally developed in order to arrive at an instrument, which would allow these distinctions to be made in reference to stereotypical gender roles, namely Femininity, Masculinity and Social Desirability.
criterion that descriptions of masculinity and femininity should describe something positive and stereotypically male or female. A personality characteristic qualified as typically masculine if independently judged by both females and males to be significantly more desirable for a man than a woman (Table 1). Similarly, a characteristic qualified as typically feminine if judged by both males and females to be significantly more desirable for a woman than for a man (Table 2).
The items relating to social desirability were culled from characteristics, which seemed to be neither typically feminine nor masculine but rather applicable to both sexes and thus neutral in terms of sex-‐appropriateness. Of these half were positive in tone and half were negative. Items were judged neutral if they were independently judged by both males and females to be no more desirable for one sex than for the other (Table 3). Note, however, that this scale mainly served an initial purpose in constructing the instrument “to insure that the inventory would not simply be tapping a general tendency to endorse socially desirable traits” (Bem, 1974, p. 156).
Table 1. Items of the BSRI Masculinity Scale (Bem, 1974; 1981). Translation into Swedish included as well as their variable labels by which they were processed.
English original Swedish translation Variable
labels
Self-‐reliant Full av självförtroende M1SJLVFR Defends own beliefs Försvarande mina övertygelser M2FORSV
Independent Oberoende M3OBERO
Athletic Sportslig M4SPORT
Assertive Bestämd M5BESTMD
Strong personality En stark personlighet M6STPER
Forceful Kraftfull M7KRFTFL
Analytical Analytisk M8ANALYT
Has leadership abilities En ledartyp M9LEDARE Willing to take risks Villig att ta risker M10RISK Makes decisions easily Lätt för att ta beslut M11BSLUT Self-‐sufficient Självtillräcklig M12SJLVT
Dominant Dominant M13DOMIN
Masculine Maskulin M14MASK
Willing to take a stand Villig att fatta beslut M15BSLST och stå för dem
Aggressive Aggressiv M16AGGRS
Act as a leader Att agera ledare M17LEDAR
Individualistic Individualist M18INDIV
Competitive Ha en tävlingsmentalitet M19TAVL
Ambitious Ambitiös M20AMBI
Table 2. Items of the BSRI Femininity Scale (Bem, 1974; 1981). Translation into Swedish included as well as their variable labels by which they were processed.
English original Swedish translation Variable labels
Yielding Undfallande F1UNDFAL
Cheerful Munter F2MUNTR
Shy Blyg F3BLYG
Affectionate Tillgiven F4TILLGV
Flatterable Mottaglig för smicker F5SMICKR
Loyal Lojal F6LOJAL
Feminine Feminin F7FEMINI
Sympathetic Sympatisk F8SYMPAT
Sensitive to the needs Känslig för andras behov F9KNSLBH of others
Understanding Förstående F10FORST
Compassionate Medlidsam F11MDLID
Eager to sooth hurt Angelägen att lindra F12LINDR feelings sårade känslor
Soft spoken Mild och stillsam F13MILD
Warm Varm person F14VARM
Tender Ömsint F15OMSIN
Gullible Lättlurad F16LTLUR
Childlike Barnslig F17BARNS
Does not use harsh Använder inte svordomar F18SVOR harsh language
Loves children Barnkär F19BARKR
Gentle Varsam F20VARSM
Table 3. Items of the BSRI Social Desirability Scale (Bem, 1974; 1981). Translation into Swedish included as well as their variable labels by which they were processed.
English original Swedish translation Variable labels
Helpful Hjälpsam SD1HJALP
Moody Lynnig SD2LYNN
Conscientious Samvetsgrann SD3SMVTG
Theatrical Teatralisk SD4TEATR
Happy Glad SD5GLAD
Unpredictable Oförutsägbar SD6OFRSB
Reliable Pålitlig SD7PLIT
Jealous Svartsjuk SD8SVART
Truthful Sanningsenlig SD9SANN
Secretive Hemlighetsfull SD10HEML
Sincere Uppriktig SD11UPPR
Conceited Inbilsk SD12INB
Likeable Trevlig SD13TREV
Solemn Allvarlig SD14ALLV
Friendly Vänlig SD15VANL
Inefficient Ineffektiv SD16INEF
Adaptable Anpassningsbar SD17ANPA
Unsystematic Osystematisk SD18OSYS
Tactful Finkänslig SD19FINK
Conventional Konventionell SD20KONV
During construction judges used a seven-‐point Likert-‐scale, ranging from one (not at all desirable) to seven (extremely desirable) when rating items. Note that the inventory outlines American stereotypes of gender roles, which are not necessarily transferable to other nations and cultures. Bem is aware of the fact that the inventory outlines feminine and masculine traits desirable to an American
population. It was emphasised during the construction of the instrument that participants should voice what they believed to be an opinion general to American society rather than uniquely their own. The psychometric properties of Bem’s original instrument are fair. Internal consistency was evaluated by allowing separate samples of men and women provide scores for both the feminine and masculine scales respectively: Both scales as scored by women show that the F-‐ scale (α = .75) as well as the M-‐scale (α = .78) have an acceptable degree of internal consistency. Similarly, both scales as scored by the male sample show an acceptable degree of consistency as well: F-‐scale (α = .87) and M-‐scale (α = .86).
Test-‐retest reliability is also acceptable, estimated in the same way with men and women providing separate scores. Test-‐retest reliability ranges from r = .76 to r = .94 (i. e. from the female sample the F-‐scale yields r = .82 and M-‐scale r = .94, whereas for the male sample F-‐scale yields r = .89 and M-‐scale r = .76).
The neutral Social Desirability Scale serves the purpose of safeguarding the conceptualisation of masculinity and femininity as separate constructs, but serves no purpose when using the completed and validated inventory. By allowing females and males separately to judge both the F-‐scale and the M-‐scale, Bem showed that sex-‐ appropriate behaviour—which is deemed to be socially desirable—as judged by either sex, was significantly higher when females judged feminine items and when males judged masculine items in comparison to when males judged feminine items and women judged masculine item. There was, however, no such significant difference when both sexes judged the neutral Social Desirability Scale. This prompted Bem
to conclude that Femininity and Masculinity are indeed separate constructs. In addition, she goes on to show that both the participating men and women are in significant agreement as to what qualifies as “sex-‐appropriate” behaviour and “sex-‐inappropriate” behaviour. In other words, when the men felt that a certain characteristic is sex-‐ inappropriate for a man or a woman the female participants would tend to agree. And conversely, when women felt that a certain characteristic is sex-‐inappropriate for a man or a woman the participating men would largely concur.
Androgyny, the measure of which is the ultimate objective of reconceptualising masculinity and femininity, reflects the relative levels of masculinity and femininity that any individual includes in their self-‐description. Androgyny is thus expressed as an index that shows the extent to which a person is sex-‐typed on the basis of how they score on the M-‐scale and F-‐scale together. The greater the value of the Androgyny Index (AI) the more the person is sex-‐typed or sex-‐ reversed. The smaller the value of AI the more an individual could be considered androgynous.
Bem has proposed three ways of calculating AI. Originally AI was defined as Student’s t - ratio for the difference between masculine and feminine self-‐endorsement. In other words AI is the difference between an individual’s Masculinity and Femininity scores normalised with respect to the standard deviations of the scores from both scales and expressed as the Student’s t - ratio. In this line of reasoning individuals are classified as sex-‐typed, masculine or feminine if the androgyny t -‐ ratio reaches statistical significance (| t | ≥ 2.2025, df = 38, p < .05). An individual is classified as androgynous if the absolute
value of the t -‐ ratio is less or equal to one. Bem points out that these cut-‐off points are somewhat arbitrary and may be adjusted if need be. Bem offers a simpler way of assessing the AI. If multiplying the difference score between the F-‐scale and the M-‐scale by a conversion factor of 2.322 one arrives at an index which correlates nigh perfectly with the t - ratio calculation (r = .98). This conversion factor was derived from the original American normative sample (n = 917), but Bem does not convey by what means. Note that this “short-‐cut” was used by, for example, Hassler (1991) in her experimental studies of spatial abilities, musical composition ability and their relationship to the Bem’s construct of psychological androgyny.
The original three-‐way classification by Bem into gender-‐role as masculine, feminine or androgynous, however, was criticised by Spence, Helmreich and Stapp (1975). They argued that classification based on Student’s t - ratio obscures one important distinction made possible by understanding femininity and masculinity as separate constructs, namely the fact that it is also possible to score high on both scales or low on both scales rather than preferring one to the other. Bem (1977) investigated the possibility to consider a four-‐way classification, and found that the critique of merely a three-‐way classification was warranted. At a later stage in the development of the inventory, therefore, as the BSRI became commercially available, the procedures for classification slightly changed. Bem (1981) now recommends that classification be done by means of a median split classifying individuals into four categories of gender role orientation: Undifferentiated, Androgynous, Feminine and Masculine. The step-‐by-‐ step procedure by which to apply the median split for classification
will be discussed below and in relation to the Swedish revision of the BSRI.
The normative samples used by Bem in developing the instrument consists of approximately 900 individuals made up of the following groups: Caucasian undergraduates (n = 32), Afro-‐American undergraduates (n = 63), Hispanic undergraduates (n = 35), Adolescents aged 14 -‐ 17 (n = 29), Age group 20 -‐ 30 (n = 108), Age group 31 -‐ 65 (n = 55), Older adults ages >62 (n = 60) and Psychiatric in-‐patients (n = 55). The averaged norms for these groups are given in Table 4 below.
The distribution of categorisations in the normative sample according to Bem’s (1974) original classification scheme, and as based on Student’s averaged t - ratios, shows that amongst males 7 % are sex-‐ typed as feminine, 7 % as near-‐feminine, 39 % as androgynous, 18 % as near-‐masculine and 29 % as masculine. Amongst females on the other hand, 37 % are sex-‐typed as feminine, 14 % as near-‐feminine, 32 % as androgynous, 9 % as near-‐masculine and 8 % as masculine.
Table 4. Raw score means, medians and standard deviations for the F- scale, M-scale and the F-minus-M difference score of the normative sample. Note that the category sexes combined has been statistically weighted in order to correct the unequal numbers of men and women making up the sample (Bem, 1981)
Females Males Sexes combined (n = 340) (n = 476) t Femininity Mean 4.82 5.05 4.59 11.95* MD 4.90 5.10 4.60 SD .59 .53 .55 Masculinity Mean 4.95 4.79 5.12 7.03* MD 4.95 4.80 5.10 SD .68 .66 .65 F-minus-M Mean -‐ .01 6.30 -‐6.33 13.09* MD .97 6.83 -‐6.50 SD 14.94 13.35 13.37 * p < .001
Table 5. Distribution of subjects in the combined normative samples into sex-role categories according to the four-way classification using the median split method (Bem, 1974)
Females Males Sex-role category (%) (%) Feminine 39 12 Masculine 12 42 Androgynous 30 20 Undifferentiated 18 27
Distribution of categories in the same sample but as based on median splits are accounted for in Table 5.
It is possible to evaluate how the two classification schemes compare if the values of near-‐feminine and near-‐masculine are combined and regarded—for the purpose of comparison—as more or less the same as being undifferentiated. In the male sample the t - ratio classification corresponds only moderately to the median split classification (r = 0.45), whereas in the female sample there is little difference between two schemes (r = 0.949).
The BSRI in a Swedish context
The nature of the Swedish sample
The Swedish sample (n = 118) is relatively small in comparison to the normative sample used by Bem. The objective of the present research, however, is mainly to investigate how the BSRI fares in a Swedish context rather than to strictly standardise the instrument anew in a fully fledged large-‐scale effort. On the other hand, Swedish norms will be produced and the present data may indicate the psychometric value of the instrument as such. At the very least this evaluation may certainly provide a base of data, which can be used to probe psychometric properties further. For this purpose the raw mean scores are appended to this report (Appendix II). A breakdown of the Swedish is provided in Table 6. Note that there is an over-‐representation of females in the sample (41 % males and 59 % females). This will need to be considered as the normative scores are produced. The same over-‐
representation problem is evident in almost exactly the same proportions in Sandra Bem’s normative sample. However, in Bem’s sample over-‐representation is reversed (42 % are female and 58 % are male).
The original American BSRI from 1974 was translated into Swedish and the administered answering sheet included all three scales (see Tables 1 -‐ 3 above for a comparison between the original American items and their translation into Swedish). A majority of the respondents filled in the translated during class and always because course content qualified or included a discussion of gender-‐roles or socialisation. The inventory was always completed prior to any discussion of gender-‐roles, however, and respondents were only provided with the information needed complete the task.
Scale items on the scoring sheets were compiled in such a manner that order effects were avoided. It was not possible for participants to perceive that in fact three scales rather than one were being rated. The order of the items was set in groups of three in the following order: Masculine, Feminine and Neutral, which is also the order used by Bem. Note that Swedish participants were encouraged to rate themselves, which is contrary to Bem’s instructions to the American sample. Bem asked that participants should express what they felt was typically American in a general sense rather than using the BSRI to provide a personal profile.
Table 6. The different subgroups of the Swedish test sample. Note that the
percentages pertaining to each sex is the relative frequency of the subgroup, whereas the relative frequency of n relates to N.
Age
____________________________
Subgroup n M F Mean Min Max SD
Pre-‐school student teachers 24 4 20 23.3 19 40 5.01 Relative frequency (%) (20) (17) (83)
Communication Studies students 49 11 38 24.7 20 44 5.21 Relative frequency (%) (42) (22) (78)
Amateur football players 12 12 -‐-‐ 39.4 25 54 9.1 Relative frequency (%) (10) (100) -‐-‐
Pre-‐school teachers 16 4 12 48.9 24 61 9.55 Relative frequency (%) (14) (25) (75)
Comprehensive student teachers 12 12 -‐-‐ 36.0 21 51 10.45 Relative frequency (%) (10) (100) -‐-‐
Various university staff 5 5 -‐-‐ 43.0 18 60 16.07 Relative frequency (%) (4) (100) -‐-‐ All subgroups (N) 118 48 70 31.2 18 6 112.00 Relative frequency (%) (100) (41) (59)
All the Swedish participants completed the answering sheets impeccably. At all occasions students were provided with immediate feedback as everyone had completed the task. That is, after completion the students themselves were allowed to score their sheets after being informed which items belonged to which scale. Scoring was straightforward and students were able to estimate themselves whether they were high -‐ high, low -‐ low or high -‐ low on either the M-‐
scale or F-‐scale. The neutral Social Desirability Scale was ignored. Participants only had access to their own score so as to keep the ethical integrity of the testing situation. No further conclusions were drawn by students except for their relative standing in comparison to the American-‐elicited items.
However, the administration tended to trigger interesting discussions on gender and the face validity of the instrument. In some cases, particularly with regard to the Communication Studies students, the inventory caused some merriment in the class. They argued that items were “out-‐dated” and not at all appropriate for outlining masculine and feminine stereotypical behaviour in Sweden.
In evaluating the BSRI in the following estimations are made on the basis of three types of scores: judgements by the men of the sample, judgements made by the women of the sample and both men and women combined. This is also the strategy chosen by Bem for producing the original norms for BSRI.
Reliability of the inventory scales
Cronbach’s Coefficient α was used as a measure of internal consistency also for the translated version of BSRI (Table 7). It would appear that reliability from this perspective largely concurs with that of Bem’s normative sample with one exception. The Social Desirability Scale as judged by the Swedish sample shows poor consistency (α = .56 for the combined sample), whereas the same scale in Bem’s original 1973 sample is acceptable (α = .70 for the combined Stanford sample). However, the SD-‐scale is nevertheless less reliable than the M-‐scale and the F-‐scale also in the Bem normative sample. After criticism from
Walkup and Abbot (1978), Bem proposed that the Social Desirability Scale should no longer play a part in the inventory. The items are still included in the inventory but serve only the purpose of being “fillers”; providing a general environment. They are rated by respondents but not considered by the research or clinician making use of the inventory. The Social Desirability Scale has been removed completely in the Swedish revision of BSRI.
The fact that the SD-‐scale is of questionable reliability in the Swedish setting strongly suggests that it should not be used to qualify or verify the independence of androgyny as a construct, which was originally done by Bem in the normative sample. She showed that androgyny was nearly uncorrelated with androgyny as expressed by Student’s t - ratio (that is r = .03 for males and r = - .10 for females).
Table 7. Coefficient alpha for both the Swedish sample and Bem’s 1978 Stanford
sample
Scale Males Females Combined
Masculinity Swedish sample .80 .82 .82 Stanford sample .86 .87 .86* Femininity Swedish sample .84 .74 .81 Stanford sample .78 .78 .80* Social Desirability Swedish sample .63 .47 0.56 Stanford sample n/a n/a .70*
Either an indigenously derived scale of Social Desirability for a Swedish context be constructed to replicate Bem’s construct procedure, or the averaged F-‐scale and M-‐scale means are correlated with each other to estimate the degree to which they correspond. A high degree of correspondence would suggest that Masculinity and Femininity as separate constructs is not a successful one, whereas a low degree of correspondence would suggest the opposite. Bem offers this reliability check also in the commercially published version of the inventory and thus shows that the two remains uncorrelated (Females r = .00 and Males r = -‐.05).
However, this is not the case for the Swedish sample. Whilst correspondence between the scales as judged by the female participants is low (r = .16) the correspondence as judged by the male participants is moderate (r = .41). The same procedure applied to the total Swedish sample, including both men and women, also signifies a degree of correspondence although weak (r = .30).
These results suggest that the use of BSRI in a Swedish setting in merely a translated form is not straightforward in spite of acceptable alpha values for both scales. There appears to be some overlap between the Masculinity Scale and the Femininity Scale, which are likely due to cultural differences. This conceptual discrepancy between the samples requires closer scrutiny.
Performing an inter-‐item total correlation analysis shows that a number of variables are either weakly correlated to the scales or that men and women strongly disagree on certain items as being sex-‐ appropriate (Table 8). For example, it appears that being athletic (M4SPORT), analytical (M8ANALYT), competitive (M19TAVL) or
ambitious (M20AMBI) is considered by both men and women not to be typically masculine. Similarly all participants, irrespective of sex, seem to agree that yielding (F1UNDFAL),
Table 8. Inter-item total correlation as based on female and male participant scores
separately and combined. To decipher item labels see Table 1 & 2 above. Items which either correlate weakly to the scale or sample to which they apply, or items regarding which men and women appear to disagree strongly, are boldened and enlarged.
Scale items (M) M F Comb Scale items (F) M F Comb M1SJLVFR .51 .50 .52 F1UNDFAL .17 -.01 .06 M2FORSV .50 .54 .53 F2MUNTR .23 .17 .25 M3OBERO .13 .61 .40 F3BLYG .22 .21 .16 M4SPORT .01 -.10 -.03 F4TILLGV .64 .48 .59 M5BESTMD .65 .52 .56 F5SMICKR .31 .31 .25 M6STPER .63 .34 .43 F6LOJAL .45 .32 .39 M7KRFTFL .53 .54 .56 F7FEMINI .36 .51 .49 M8ANALYT .15 .32 .27 F8SYMPAT .57 .40 .50 M9LEDARE .39 .58 .59 F9KNSLBH .64 .39 .56 M10RISK .33 .53 .46 F10FORST .48 .38 .48 M11BSLUT .42 .39 .44 F11MDLID .61 .53 .58 M12SJLVT .53 .31 .35 F12LINDR 55 .42 .52 M13DOMIN .58 .56 .55 F13MILD .41 .46 .41 M14MASK .39 .34 .38 F14VARM .51 .54 .57 M15BSLST .39 .58 .53 F15OMSIN .73 .61 .70 M16AGGRS .37 .09 .21 F16LTLUR .40 -.05 .17 M17LEDAR .39 .48 .48 F17BARNS .25 .07 .15 M18INDIV .14 .55 .34 F18SVOR .12 .22 .20 M19TAVL .21 .25 .29 F19BARKR .41 .23 .38 M20AMBI .32 .23 .21 F20VARSM .72 .42 .59
cheerful (F2MUNTR), shy (F3BLYG), flatterable (F5SMICKR) and childlike (F17BARNS) do not describe characteristics that are typically feminine, nor is use or no use of harsh language (F18SVOR) a variable
appropriate to describe femininity. Observe, however, that men disagrees with women on the status of being independent (M3OBERO), aggressive (M16AGGRS) and individualistic (M18INDIV). The women suggest that independence and individualism are typical male characteristics whereas men do not. On the other hand, men feel that aggression to some degree is typically male while women appear to disagree. In a like manner men and women do not agree on whether women are typically gullible (F16LTLUR). Men say they are while women object. Also, men seem to think, to a higher degree than do women, that it is typically feminine to love children (F19BARKR).
In order to come to terms with the response patterns of the Swedish sample—deviant in comparison to the American normative sample—and increase the integrity of the two scales, seven items were removed from the Masculinity Scale (M3, M4, M8, M16, M18, M19 and M20) and eight items from the Femininity Scale (F1, F2, F3, F5, F16, F17, F18, F19). A new reliability analysis was performed on the resulting shortened version of the translated inventory.
This procedure increased internal consistency as expressed by Coefficent Alpha for both scales and with regard to females, males and men and women combined (Table 9). Inter-‐item correlations have also become more robust (Table 10). To further confirm the conceptual basis of the BSRI in a Swedish setting a factor analysis (principal components with varimax rotation) was performed on the combined and revised scales. A first run revealed a solution of seven factors explaining 66.5 % of total variance. However, in this solution two factors dominate and together explain 41.4 % of total variance. The variance of the remaining five factors ranges from 4.0 to 7.3 %. In a
second run the principal component analysis was restricted to a two-‐ factor solution only, which would seem to corroborate the existence of two separate constructs (Table 11) explained by approximately equal proportions of the total variance (22.0 % and 19.3 % respectively). However, there is still some overlap between the two scales. Masculinity (M14MASK) loads moderately on both factors and the same is true of Soft-‐spoken (F13MILD).
Table 9. Coefficient alpha for both the unrevised and the revised Swedish version of
BSRI Masculinity Femininity ______________________ _______________________
Sample Unrevised Revised t Unrevised Revised t Males .80 .85 33.0* .84 .87 57.0* Females .82 .84 83.0* .74 .83 17.4 Combined .82 .84 83.0* .81 .86 33.4* * p< .01
Table 10. Inter-item total correlation as based on the corrected Femininity and
Masculinity Scales. To decipher item labels see Table 1 & 2 above
Scale items (M) M F Combined Scale items (F) M F Combined M1SJLVFR .43 .54 .51 F4TILLGV .59 .58 .60 M2FORSV .51 .46 .48 F6LOJAL .43 .34 .39 M5BESTMD .66 .52 .57 F7FEMINI .39 .44 .49 M6STPER .64 .40 .47 F8SYMPAT .58 .46 .52 M7KRFTFL .52 .56 .57 F9KNSLBH .65 .56 .65 M9LEDARE .59 .63 .64 F10FORST .58 .49 .57 M10RISK .40 .47 .44 F11MDLID .61 .57 .59 M11BSLUT .47 .45 .49 F12LINDR .65 .46 .58 M12SJLVT .48 .30 .32 F13MILD .32 .37 .32 M13DOMIN .58 .58 .56 F14VARM .55 .58 .61 M14MASK .42 .30 .37 F15OMSIN .77 .71 .76 M15BSLST .43 .54 .51 F20VARSM .73 .47 .61 M17LEDAR .43 .53 .52